The protocol appears to be operating correctly in the "gettingafilefromamigaatawin2" capture. The slowdown is the large time gaps between the receiving end (192.168.1.10) sending an acknowledgement, and the sending end (192.168.1.120) sending the next data packet. The gaps at that point appear to frequently be as much as two seconds. It looks like the sender is busy doing something other than sending. Strangely, this seems to occur most often near the end of the transmission of a 64K chunk. This pause is absent during the transmission of the last chunk. I wonder if the sender is busy in the middle of the transfer with queuing up the next chunk. Slow disk access?
-----Original Message-----
From: Ulf Bertilsson [mailto:[EMAIL PROTECTED]]
Sent: Monday, June 24, 2002 10:20 AM
To: Esh, Andrew; [EMAIL PROTECTED]; [EMAIL PROTECTED]
Cc: [EMAIL PROTECTED]
Subject: RE: Very bad performance when copying large files from windows
to samba-share
Importance: Low
Well, if your still feel up for it :)
Amithlon(AmigaOS on x86 layer)->Win2k (800kb/s)
Win2k<-Amithlon(AmigaOS on x86 layer) (20-40Kb/s)
I could provide an debug level 9 output.
Exotic things..
Find the NBSS checksum errors :) (Parser error in Ethereal ?)
Also look how strange behaviour of windows size when you "get" a file
from Amithlon.
17520
17520
...
16...
16...
Seems unsynched somehow ?
The other way things seems smother with an stable windows size of 7300,
like:
tcp
NBSS
NBSS
tcp
...
http://www.birrabrothers.com/tiger/download/gettingafilefromamigaatawin2
kbox.capture
Will reflect this, could be I'm confused with some basic tcp/ip thing..
But if your up to it =D
Amiga port of smbfs works neat tho with high rates and you can even
star/stop
the interface in realtime, reboot other host etc =D
So I think it's isolated to an Samba issue.
Should I keep the list out of the thread now ?
--
Ulf
-----Original Message-----
From: Esh, Andrew [mailto:[EMAIL PROTECTED]]
Sent: Monday, June 24, 2002 4:42 PM
To: Ulf Bertilsson; Esh, Andrew; [EMAIL PROTECTED];
[EMAIL PROTECTED]
Cc: [EMAIL PROTECTED]
Subject: RE: Very bad performance when copying large files from windows
to samba-share
I looked at the captures of win98 uploading to amithlon and redhat. They
are not rabbit-pellet mode. The protocol is using full sized packets,
64K chunks, and has between two and four packets on the wire at a time.
There are no flush commands present, which is the main characteristic of
the slow transfer mode.
Those two transfers appear to be operating rather quickly, with
end-to-end response time in the 6-10ms range, and streamed packet
intervals around 30us. I measured one 64K chunk taking 260ms to be
transferred, which is close to a 2Mb/S transfer rate.
Is there something else I should look at?
-----Original Message-----
From: Ulf Bertilsson [mailto:[EMAIL PROTECTED]]
Sent: Monday, June 24, 2002 2:54 AM
To: Esh, Andrew; [EMAIL PROTECTED]; [EMAIL PROTECTED]
Cc: [EMAIL PROTECTED]
Subject: RE: Very bad performance when copying large files from windows
to samba-share
Importance: Low
I finaly got the mashine up and running again =D
If it's still interesting here is a download point for captures..
http://www.birrabrothers.com/tiger/download/
I'm puzzeled over that I also have this error in win2k.
--
Ulf
-----Original Message-----
From: Esh, Andrew [mailto:[EMAIL PROTECTED]]
Sent: Monday, April 22, 2002 4:15 PM
To: Ulf Bertilsson; [EMAIL PROTECTED]; [EMAIL PROTECTED]
Cc: [EMAIL PROTECTED]
Subject: RE: Very bad performance when copying large files from windows
to samba-share
Please send a copy of any tcpdumps to me as well, if you would. I spent
a lot of time characterizing the Rabbit Pellet transfer mode. I'll
recognize it right away, if that is what this is.
-----Original Message-----
From: Ulf Bertilsson [mailto:[EMAIL PROTECTED]]
Sent: Monday, April 22, 2002 3:01 AM
To: [EMAIL PROTECTED]; [EMAIL PROTECTED]
Cc: [EMAIL PROTECTED]
Subject: RE: Very bad performance when copying large files from windows
to samba-share
Importance: Low
> > Just out of curiosity, what did security=server contribute:
> > added latency in some obscure part of the process???
>
> As far as we could tell... yes. ;)
>
> I really should have documented that whole episode.
I have an current "issue" with Samba 2.0.7 and an Win98 box.
Peak at ~800kb/s one way, 20-30kb/s other way (10MB enviroment)
I have some captures I will show Chris.
Could be relevant in this contex..
Otherwise, sorry for the OT :)
--
Ulf
